From 1659fb9b43efaf717152d7e5296a578984bbde50 Mon Sep 17 00:00:00 2001
From: Andrew Gallant <jamslam@gmail.com>
Date: Sat, 30 Sep 2023 14:16:51 -0400
Subject: [PATCH] printer: hand-roll decimal formatting

It seems like a trifle, but if the match frequency is high enough, the
allocation+formatting of line numbers (and columns and byte offsets)
starts to matter. We squash that part of the profile in this commit by
doing our own decimal formatting. I speculate that we get a speed-up
from this by avoiding the formatting machinery and also a possible
allocation.

An alternative would be to use the `itoa` crate, and it is indeed
marginally faster in ad hoc benchmarks, but I'm satisfied enough with
this solution.
---
 crates/printer/src/hyperlink.rs |  6 ++--
 crates/printer/src/standard.rs  |  8 ++---
 crates/printer/src/util.rs      | 60 +++++++++++++++++++++++++++++++++
 3 files changed, 67 insertions(+), 7 deletions(-)

diff --git a/crates/printer/src/hyperlink.rs b/crates/printer/src/hyperlink.rs
index 34ba5914..279258c1 100644
--- a/crates/printer/src/hyperlink.rs
+++ b/crates/printer/src/hyperlink.rs
@@ -5,7 +5,7 @@ use {
     termcolor::{HyperlinkSpec, WriteColor},
 };
 
-use crate::hyperlink_aliases;
+use crate::{hyperlink_aliases, util::DecimalFormatter};
 
 /// Hyperlink configuration.
 ///
@@ -484,11 +484,11 @@ impl Part {
             ),
             Part::Path => dest.extend_from_slice(&values.path.0),
             Part::Line => {
-                let line = values.line.unwrap_or(1).to_string();
+                let line = DecimalFormatter::new(values.line.unwrap_or(1));
                 dest.extend_from_slice(line.as_bytes());
             }
             Part::Column => {
-                let column = values.column.unwrap_or(1).to_string();
+                let column = DecimalFormatter::new(values.column.unwrap_or(1));
                 dest.extend_from_slice(column.as_bytes());
             }
         }
diff --git a/crates/printer/src/standard.rs b/crates/printer/src/standard.rs
index c33bf14e..48729974 100644
--- a/crates/printer/src/standard.rs
+++ b/crates/printer/src/standard.rs
@@ -24,7 +24,7 @@ use crate::{
     stats::Stats,
     util::{
         find_iter_at_in_context, trim_ascii_prefix, trim_line_terminator,
-        PrinterPath, Replacer, Sunk,
+        DecimalFormatter, PrinterPath, Replacer, Sunk,
     },
 };
 
@@ -1709,7 +1709,7 @@ impl<'a, M: Matcher, W: WriteColor> PreludeWriter<'a, M, W> {
     fn write_line_number(&mut self, line: Option<u64>) -> io::Result<()> {
         let Some(line_number) = line else { return Ok(()) };
         self.write_separator()?;
-        let n = line_number.to_string();
+        let n = DecimalFormatter::new(line_number);
         self.std.write_spec(self.config().colors.line(), n.as_bytes())?;
         self.next_separator = PreludeSeparator::FieldSeparator;
         Ok(())
@@ -1723,7 +1723,7 @@ impl<'a, M: Matcher, W: WriteColor> PreludeWriter<'a, M, W> {
         }
         let Some(column_number) = column else { return Ok(()) };
         self.write_separator()?;
-        let n = column_number.to_string();
+        let n = DecimalFormatter::new(column_number);
         self.std.write_spec(self.config().colors.column(), n.as_bytes())?;
         self.next_separator = PreludeSeparator::FieldSeparator;
         Ok(())
@@ -1736,7 +1736,7 @@ impl<'a, M: Matcher, W: WriteColor> PreludeWriter<'a, M, W> {
             return Ok(());
         }
         self.write_separator()?;
-        let n = offset.to_string();
+        let n = DecimalFormatter::new(offset);
         self.std.write_spec(self.config().colors.column(), n.as_bytes())?;
         self.next_separator = PreludeSeparator::FieldSeparator;
         Ok(())
diff --git a/crates/printer/src/util.rs b/crates/printer/src/util.rs
index b633ec9a..2d20a7df 100644
--- a/crates/printer/src/util.rs
+++ b/crates/printer/src/util.rs
@@ -397,6 +397,47 @@ impl Serialize for NiceDuration {
     }
 }
 
+/// A simple formatter for converting `u64` values to ASCII byte strings.
+///
+/// This avoids going through the formatting machinery which seems to
+/// substantially slow things down.
+///
+/// The `itoa` crate does the same thing as this formatter, but is a bit
+/// faster. We roll our own which is a bit slower, but gets us enough of a win
+/// to be satisfied with and with pure safe code.
+#[derive(Debug)]
+pub(crate) struct DecimalFormatter {
+    buf: [u8; Self::MAX_U64_LEN],
+    start: usize,
+}
+
+impl DecimalFormatter {
+    /// Discovered via `u64::MAX.to_string().len()`.
+    const MAX_U64_LEN: usize = 20;
+
+    /// Create a new decimal formatter for the given 64-bit unsigned integer.
+    pub(crate) fn new(mut n: u64) -> DecimalFormatter {
+        let mut buf = [0; Self::MAX_U64_LEN];
+        let mut i = buf.len();
+        loop {
+            i -= 1;
+
+            let digit = u8::try_from(n % 10).unwrap();
+            n /= 10;
+            buf[i] = b'0' + digit;
+            if n == 0 {
+                break;
+            }
+        }
+        DecimalFormatter { buf, start: i }
+    }
+
+    /// Return the decimal formatted as an ASCII byte string.
+    pub(crate) fn as_bytes(&self) -> &[u8] {
+        &self.buf[self.start..]
+    }
+}
+
 /// Trim prefix ASCII spaces from the given slice and return the corresponding
 /// range.
 ///
@@ -527,3 +568,22 @@ where
     dst.extend(&bytes[last_match..end]);
     Ok(())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn custom_decimal_format() {
+        let fmt = |n: u64| {
+            let bytes = DecimalFormatter::new(n).as_bytes().to_vec();
+            String::from_utf8(bytes).unwrap()
+        };
+        let std = |n: u64| n.to_string();
+
+        let ints = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 20, 100, 123, u64::MAX];
+        for n in ints {
+            assert_eq!(std(n), fmt(n));
+        }
+    }
+}